Compute VPC quantile statistics from preprocessed data

Internal function. Always emits both standard and prediction-corrected summary statistics: the std flavor uses raw OBSDV/SIMDV; the pc flavor applies var_loqcens to SIMDV (when loq is supplied) and var_predcorr() per (bin_var × strat_var × CMT) to both OBSDV and SIMDV. The pc columns are returned with a pc_ prefix.

Usage

df_vpccompute(
  data,
  pi = c(0.05, 0.95),
  ci = 0.9,
  bin_var = BIN_MID_VAR,
  strat_var = NULL,
  irep_name = "SIM",
  lower_bound = 0,
  mode = c("auto", "rank", "drop"),
  loq = attr(data, "loq")
)

Arguments

data: Preprocessed simulation data, typically from df_vpcpreprocess().
pi: Numeric vector of length 2 specifying prediction interval quantiles. Default is c(0.05, 0.95).
ci: Numeric scalar in (0, 1) specifying the simulation interval (e.g. 0.90 for a 90% CI). Default is 0.90.
bin_var: String. Binning variable name. Default is "BIN_MID".
strat_var: String or NULL. Stratification variable name. Default is NULL.
irep_name: String. Replicate identifier column name. Default is "SIM".
lower_bound: Numeric. Forwarded to var_predcorr() in the pc flavor.
mode: One of "auto" (default), "rank", or "drop". Controls how BLQ-encoded values flow into quantile aggregation per-flavor: "auto" uses "rank" for the std flavor and "drop" for the pc flavor (the historical defaults); "rank" and "drop" apply to both flavors.
loq: Numeric vector of unique LOQ values present in data (length 1 when LLOQ is constant), or NULL. Defaults to attr(data, "loq"). When non-NULL, the std stats include sim_prop_blq_low/med/hi (BLQ detection reads per-row thresholds from data$LOQ). The pc stats do not emit a pc_sim_prop_blq_* set — LOQ has no meaning on the prediction-corrected scale.

Value

A list of two data.frames with class c("vpc_stats", "list"):

stats: Wide summary with both std and pc columns. Std columns are unprefixed (obs_low/med/hi, sim_low_* etc.); pc-flavor observed and simulated quantile columns carry a pc_ prefix (pc_obs_low, pc_sim_low_med, etc.). obs_n, obs_n_blq, obs_prop_blq, sim_prop_blq_*, ci, pi_low, pi_hi are single (not duplicated). Carries attributes n_replicates, loq, and strat_var.
obs: First-replicate observation rows with MDV == 0, carrying both OBSDV (std) and PC_OBSDV (prediction-corrected).